Nicolas Bred and #xE8;che and Jean-Daniel Zucker
This paper describes an approach to ground names of objects seen by a mobile robot in a real world environment. Names are given by a human teacher and axe stored within a lexicon, along with images where the corresponding objects are present. Here, the building of a high-level perceptual system is required to enable an efficient learning of the physical definition of objects. The paper briefly describes how an abstraction-based reformulation of real world data can highlight relevant information for a given concept.